SandForce Attacks SSD Limitations

Many enterprises have been reluctant to adopt solid-state disk drive (SSD) technology, worried about a variety of issues such as quality of service with write performance, overall product durability, and high cost. This comes at a time when industry researcher Enterprise Strategy Group reports that a survey of enterprise IT managers indicates that 47 percent think SSDs are too expensive when compared to hard-disk drives and 48 percent say that they are not interested because SSDs have either limited drive capacity or no applications where flash makes sense. Another 19 percent say they're worried about reliability and longevity when it comes to flash-based SSDs.

A new vendor SandForce recently emerged from stealth mode to unveil the SF-1000 SSD Processor Family, a replacement for conventional storage controllers that it says will address enterprise concerns about flash SSD endurance, reliability, and data retention.

Other suppliers have tried to answer enterprise flash SSD challenges by focusing on product buildouts at lower geometries in order to reduce production (and therefore retail) costs. They've also incorporated wear-leveling algorithms to extend flash SSD lifespans and are using integrated, automated storage management to reserve solid-state disk for high-demand processing and spreading the work of the enterprise over a mix of solid-state and hard-drive devices so as to utilize "best choice" storage media.

"There's both fear and trepidation in the minds of enterprise storage managers when it comes to moving from hard drives to SSDs," says Jim Handy, semiconductor analyst for research firm Objective Analysis. "A transition to solid state technology is made that much easier if these managers can just do a one-to-one replacement of a hard drive with an SSD. It is easy to see how SSD performs in this context, and to make a decision as to whether SSD is providing value or it isn't. If you do this on a small scale, you can plug in one SSD and replace three hard drives, and you'll see a modest improvement with two empty hard drive slots."

The approach is reassuring for many enterprise storage managers who lack practical experience with SSDs, are just getting their feet wet, and are fully cognizant of the career risks of being first to deploy a new technology.It is also normal during this evaluative process to look to storage vendors for guidance. But the variety of vendor wear-leveling algorithms, coupled with warranties that prescribe SSD media use for a certain number of writes, still do not cover all of the enterprise concerns about SSD cost, durability and performance.

SandForce is trying to deal with those issues in a different way. "If you want to overcome enterprise objections and really replace hard drives, the SSD solution has to look and be like a hard drive," says Alex Naqui, the company's president and CEO. "The challenge is that as geometries shrink to make NAND flash less expensive, endurance, reliability, and data retention also drop. If you really want to compete with a hard drive, you have to be able to warranty the lifetime of a NAND flash solution for the five years that enterprises routinely expect out of their hard drives and not to varying time spans of different vendors' warranties for flash. You also have to present a flash solution that does not come out of the box to the enterprise with a warning that only so many writes should be performed over a specific period of time. It is only after these issues are resolved that we are going to get enterprises to broadly adopt the product."

Ideally, SandForce sees the flash SSD implementation of choice as commoditized MLC, because MLC is much cheaper than SLC and will contribute positively to the kinds of cost reductions that enterprises want to see. On the non-commodity side of the equation, the SandForce offering builds proprietary technology into a chip that SandForce feels will eliminate enterprise performance, capacity, data retention, and durability objections.

SandForce talks about its DuraClass technology, which it says can enable flash to be an enterprise success. DuraClass takes aim at eliminating the SSD performance, endurance, and reliability concerns by optimizing the number of flash program cycles to extend flash endurance by 80x when compared to standard controllers. It also implements a proprietary flash media error-correction methodology known as RAISE (redundant array of independent silicon elements), which SandForce says will deliver orders-of-magnitude improvement in drive reliability over both SSDs and HDDs. It also provides advanced read and program disturb management intelligence, designed to safeguard against errant reprogramming of memory cells and unexpected power loss.

SandForce tackles the cost concerns of enterprises by making its controller technology entirely memory-agnostic. In other words, the SandForce technology can work with any SLC or MLC memory from any supplier, making NAND memory a commodity item amenable to least-cost implementation. To round out the proposition, this controller and memory on a chip works with any SATA (serial advanced technology attachment) or third-party SAS (serial attached SCSI) interfaces."It's too early to say, but if their approach works it could be unique," says John Chen, senior director at research firm TrendFocus. "They have to go through proofs of concept first, most likely starting with SLC. We don't know how that will work out, but we do know that they have good venture funding, that some major OEMs that are partnering with them, and that they have the money to get through these initial phases of qualification, so there could be something there."

According to Handy, SandForce has sampled out a prototype system, and approximately 10 to 20 systems are now under evaluation with different companies. If the trials prove out, it might prompt many flash SSD suppliers to adopt the SandForce technology, which in turn might solve some of the inherent problems that suppliers have experienced with their own controllers -- and put the product into the enterprise marketplace in 2010.

What SandForce is saying will ultimately be validated by what it can deliver,” says Greg Wong, analyst with Forward Insights. "The first step is proving out the product with SLCs and working with partners, where there likely will be adjustments made to firmware by the companies working with the product. It won't be like simply turning on a switch, because there is always fine tuning. However, if SandForce can achieve its objectives, this could be a game-changer for flash SSD. For now, we have to wait and see."

InformationWeek Analytics has published an independent analysis of the challenges around enterprise storage. Download the report here (registration required).